Mask Detection Method Based on YOLO-GBC Network

For the problems of inaccurate recognition and high missed detection rate existing mask algorithms in actual scenes, a novel algorithm based on YOLO-GBC network is proposed. Specifically, backbone part, global attention mechanism (GAM) integrated to improve ability extract key information through cross-latitude interaction. The cross-layer cascade method adopted feature pyramid structure achieve effective bidirectional cross-scale connection weighted fusion. sampling content-aware reassembly features (CARAFE) into fully retain semantic map. NMS replaced with Soft-NMS model prediction frame accuracy by confidence decay method. experimental results show that average (mAP) reached 91.2% data set, which 2.3% higher than baseline YOLOv5, speed 64FPS. recall have also been improved varying degrees, increasing task correctly wearing masks.